奥赖恩：教导语言模型以思维语言高效推理

ORION: Teaching Language Models to Reason Efficiently in the Language of Thought

November 28, 2025

作者: Kumar Tanmay, Kriti Aggarwal, Paul Pu Liang, Subhabrata Mukherjee

cs.AI

摘要

大型推理模型在数学、代码生成和任务规划方面表现出色，但其依赖冗长的思维链会导致高延迟、冗余和推理路径不连贯。受"思维语言假说"启发——该假说认为人类推理基于名为"心理语"的符号化、组合性心智语言——我们提出了一种训练模型进行紧凑式推理的框架。心理语将抽象推理编码为超压缩的结构化标记，使模型能以更少步骤解决复杂问题。为提升效率与准确性，我们提出"短长度偏好优化"方法，通过强化学习奖励简洁正确的解法，同时保留必要时展开长推理的能力。应用于心理语对齐模型后，该方法在保持详细推理优势的同时，实现了远高于传统方法的压缩率。在AIME 2024/2025、MinervaMath、OlympiadBench、Math500和AMC等基准测试中，ORION模型的推理标记数减少至1/4-1/16，推理延迟降低达5倍，训练成本较DeepSeek R1蒸馏模型降低7-9倍，同时保持其90-98%的准确率。ORION模型在实现2倍压缩的同时，准确率较Claude和ChatGPT-4o最高提升5%。这些结果表明，心理语式压缩推理向类人认知效率迈进了一步，可在不牺牲准确性的前提下实现实时、高性价比的推理。

English

Large Reasoning Models (LRMs) achieve strong performance in mathematics, code generation, and task planning, but their reliance on long chains of verbose "thinking" tokens leads to high latency, redundancy, and incoherent reasoning paths. Inspired by the Language of Thought Hypothesis, which posits that human reasoning operates over a symbolic, compositional mental language called Mentalese, we introduce a framework that trains models to reason in a similarly compact style. Mentalese encodes abstract reasoning as ultra-compressed, structured tokens, enabling models to solve complex problems with far fewer steps. To improve both efficiency and accuracy, we propose SHORTER LENGTH PREFERENCE OPTIMIZATION (SLPO), a reinforcement learning method that rewards concise solutions that stay correct, while still allowing longer reasoning when needed. Applied to Mentalese-aligned models, SLPO yields significantly higher compression rates by enabling concise reasoning that preserves the benefits of detailed thinking without the computational overhead. Across benchmarks including AIME 2024 and 2025, MinervaMath, OlympiadBench, Math500, and AMC, our ORION models produce reasoning traces with 4-16x fewer tokens, achieve up to 5x lower inference latency, and reduce training costs by 7-9x relative to the DeepSeek R1 Distilled model, while maintaining 90-98% of its accuracy. ORION also surpasses Claude and ChatGPT-4o by up to 5% in accuracy while maintaining 2x compression. These results show that Mentalese-style compressed reasoning offers a step toward human-like cognitive efficiency, enabling real-time, cost-effective reasoning without sacrificing accuracy.